Skip to main content

Tasks, Reports & Proposals

Agents on your mesh don't just do what you tell them — they also surface their own work for your review. That happens through three connected artifacts: tasks, reports, and proposals.

Tasks​

A task is a unit of work assigned to an agent. Tasks come from two places:

  • You — via the dashboard or mesh_instruct.
  • Other agents — supervisors and builders can create tasks when their loops detect work that needs doing.

Each task tracks its lifecycle: created, dispatched to a node, in progress, completed (or failed). You see the full task board in the dashboard, filterable by node, creator, status, and priority.

Priority and organization​

Tasks carry a priority and optional labels so you can group related work. The mesh can auto-organize large task backlogs by topic, so a hundred individual tasks collapse into a handful of thematic clusters.

Dispatch​

When a task is ready to be worked on, the mesh dispatches it to a node. You can route manually ("send this to alice's node") or let the mesh pick based on availability and assignment.

Reports​

A report is an agent's summary of something it did or observed. Supervisors produce reports on a schedule; workers produce reports when they finish significant work or encounter notable events.

Reports are the main way a mesh that ran all night tells you what happened. The dashboard lists reports chronologically, each one expandable to the agent's full write-up, with links back to the sessions, files, and proposals it references.

Proposals​

A proposal is an agent's suggestion for work that should happen — a refactor it noticed, a bug it spotted, a dependency it wants to upgrade. Proposals are the primary way the mesh stays proactive without going rogue.

The proposal flow:

  1. An agent (usually a supervisor loop) creates a proposal with a title, body, and supporting context.
  2. The proposal lands in the Proposals panel on the dashboard.
  3. You review it — approve, reject, ask for more detail, or reassign.
  4. On approval, the proposal becomes a task and is dispatched to a node.

Why proposals matter​

Proposals are the "human in the loop" for autonomous agent behavior. They let the mesh notice things without acting on them unprompted — and they give you an audit trail of every autonomous decision the mesh wanted to make, whether or not it was approved.

Real-time updates​

The Proposals panel updates live — new proposals, status changes, and dispatches appear without a refresh.

Putting it together​

A realistic workflow:

  1. A supervisor loop runs overnight and notices your users table has no index on created_at.
  2. The supervisor creates a proposal: "Add index on users(created_at) — three queries in /api/audit would benefit."
  3. In the morning, you see the proposal, approve it.
  4. The mesh converts the proposal into a task and dispatches it to a worker.
  5. The worker agent opens a branch, writes the migration, runs tests, commits, and posts a report summarizing the change.

Nothing shipped without your approval, but you didn't have to notice the missing index — the mesh did.

Next​

  • Loops — The scheduled behaviors that generate most proposals
  • Observe — See agent activity as it happens